Transfer Learning Strategies for Jet Engine Health Monitoring with Full Flight Data
Monitoring the condition of engines is viewed as an instrumental technology to reduce costs
related to maintenance, repair, and overhaul while enhancing aircraft safety and availability
[1, 2]. By analyzing in-flight measurements from engine condition monitoring systems, we can
estimate the current health status of the aircraft engine, paving the way for efficient maintenance
strategies. Beyond observing and trending prolonged wear and tear, these monitoring tools can
detect, pinpoint, and recognize individual faults [3, 2].
In addition to present monitoring systems which rely on steady-state snapshots from flights
[4, 5, 6, 7], MTU is developing a complementary system that leverages Full Flight Data (FFD)
[2]. This enhancement doesn’t indicate a shift away from snapshot-based monitoring, which will
continue to play a relevant role in the foreseeable future. The FFD approach, while promising,
presents both opportunities and challenges. Its comprehensive nature permits the detection of
engine faults within a single flight, offering valuable insights. However, managing this influx
of data calls for sophisticated algorithms and consequently, requires increased research and
resources in data processing and analysis [8, 2].
Current approach is to train individual neural networks for each engine, even if they belong to
the same engine type. As each engine experiences a multitude of operational conditions, the
training data might not encapsulate the full spectrum of potential feature variations.
This thesis introduces the concept of a generic model trained on FFD from various engines of
the same type. This ensures comprehensive coverage of operational conditions. This generic
model will then serve as a baseline for training individual engine models for newly operated
engines. The primary objective is to not only provide comprehensive coverage but also to reduce
the uncertainties intrinsic to engine modeling, ultimately enhancing prediction quality.
The main objectives of this thesis include:
• Establishing the feasibility and implementation of a generic model.
• Fine-tuning this model to account for engine-specific effects.
• Comparing this approach to current procedures.
• Understanding the criteria differentiating a varied set of flights from a non-varied one,
considering the current reliance on consecutive flights.
• Aiming to reduce both training time and volume of training data needed for new engines,
all while increasing accuracy.
To achieve these objectives, the following proposed steps will be undertaken:
1. Preprocess FFD to ensure data integrity and quality.
2. Conduct an in-depth analysis of FFD to identify critical features for engine modeling,
especially operating conditions and environmental aspects.
3. Commence training of the generic model using the preprocessed data.
4. Apply transfer learning methodologies to adapt the model for individual engines, each
with its unique thermodynamic characteristics.
5. Evaluate and assess the generic model’s performance against set criteria.
References
[1] IATA, “Airline maintenance cost executive commentary,” IATA Montreal, QC, Canada,
2016.
[2] M. Weiss, S. Staudacher, J. Mathes, D. Becchio, and C. Keller, “Uncertainty Quantification
for Full-Flight Data Based Engine Fault Detection with Neural Networks,” Machines,
vol. 10, p. 846, Sept. 2022.
[3] Fentaye, Baheta, Gilani, and Kyprianidis, “A Review on Gas Turbine Gas-Path Diagnostics:
State-of-the-Art Methods, Challenges and Opportunities,” Aerospace, vol. 6, p. 83, July
2019.
[4] A. D. Fentaye, V. Zaccaria, and K. Kyprianidis, “Aircraft Engine Performance Monitoring
and Diagnostics Based on Deep Convolutional Neural Networks,” Machines, vol. 9, p. 337,
Dec. 2021.
[5] A. D. Fentaye, S. I. Ul-Haq Gilani, A. T. Baheta, and Y.-G. Li, “Performance-based fault
diagnosis of a gas turbine engine using an integrated support vector machine and artificial
neural network method,” Proceedings of the Institution of Mechanical Engineers, Part A:
Journal of Power and Energy, vol. 233, pp. 786–802, Sept. 2019.
[6] J. L. Pérez-Ruiz, Y. Tang, and I. Loboda, “Aircraft Engine Gas-Path Monitoring and Diagnostics
Framework Based on a Hybrid Fault Recognition Approach,” Aerospace, vol. 8,
p. 232, Aug. 2021.
[7] H. Lipowsky, S. Staudacher, M. Bauer, and K.-J. Schmidt, “Application of Bayesian Forecasting
to Change Detection and Prognosis of Gas Turbine Performance,” Journal of Engineering
for Gas Turbines and Power, vol. 132, Dec. 2009.
[8] V. E. Badea, A. Zamfiroiu, and R. Boncea, “Big Data in the Aerospace Industry,” Informatica
Economica, vol. 22, pp. 17–24, Mar. 2018.